confounding

Question 1

When does positive or negative confounding happen?

Answer 1

positive confounding (strengthening effect) happens when adjusted RR < crude RR. The confounder pulls the association away from the null.

negative confounding (attenuation effect) happens when adjusted RR > crude RR. The confounder pushes the association towards the null.

Question 2

T/F: Confounding is not a matter of statistical significance

Answer 2

Ture. We should not judge confounding by statistical comparison of crude and adjusted estimates

Question 3

4 criteria for a potential confounder

Answer 3

1. Associated with the Exposure (Levels of confounder differ between exposure groups )
2. Associated with the outcome (Confounder causally associated with outcome)
3. Associated with outcome independent of exposure (have to be associated with outcome in both E and non-E)
4. Not intermediate to the causal pathway (not a mechanism for exposure to outcome)

Question 4

Interpret the confounding result

Answer 4

After adjusting for smoking, the association between xx and xxx remained materially unchanged/ is substantially reduced. Smoking is/is not likely a confounder of the association between xx and xxx.

Question 5

Possible reasons for residual confounding (4)

Answer 5

1. incomplete assessment of confounding factor
2. Within-stratum confounding
3. Other factors related to confounder, exposure and/or outcome are operating through the confounder (unmeasured or indirect confounding) and may not fully be accounted for by adjusting for the confounder under study
4. Imperative to understand biologic framework for E-O

Question 6

Wehn dealing with confounding, we need to consider?

Answer 6

1. study question & design
2. feasibility of possible approaches
3. often try different approaches and look for consistency or inconsistency of findings

Question 7

Definition of confounding

Answer 7

Confounding happens when all or part of an observed association between exposure and disease is explained (accounted for) an extraneous factor.

Question 8

What are some approaches to handle confounding?

Answer 8

1. In designing and conducting the study
   * Restriction (exclusions)
   e. g., if smoking is a likely confounder, enroll only nonsmokers.

Issues: feasibility of enrollment; association with other

relevant study factors of interest; external validity (e.g. smokers may have social or lifestyles we are interested in)

• Randomization: balance the confounding and believe confounding had happened
• Matching

2. In the analysis of data

• Stratification (direct standardization; Mantel-Haenszel)
• Multivariable regression adjustment

Question 9

What is residual confounding

Answer 9

Effect of confounder persists on the E-O association because of incomplete adjustment

Question 10

How to do the data-specific assessment to identify potential confounders?

Answer 10

Step 1: Compare distribution of potential confounder(s) between E and non-E, and between D+ and D-

Step 2: Compare crude with adjusted measure of association (RR or OR)

• Stratified analysis

Mantel-Haenszel procedure (E-D association, stratified on confounder categories)

• if stratum ORs are uniform with each other, but*
• different from crude OR, then confounding likely exists.*
• Multivariable regression analysis

Question 11

What’s the similarity and difference between confounding and bias?

Answer 11

The effect of confounding is similar to that of bias, that is move association in one direction, either towards or away from null. However, unlike bias, which is due to error, confounding is a real phenomenon and reflects the complexity of biologic pathways of disease. We try to avoid/control bias (error), try to understand confounding

Question 12

How do we identify potential confounders?

Answer 12

1. Knowledge of the biology of the disease & exposure
2. Past experience in a similar study you conducted
3. Published findings from other similar studies
4. Data-specific assessment